This week, I confirmed one thing once again:
Accurate forecasts do not necessarily mean enough green power.
A multi-site retail customer asked us this week:
If AI can already predict electricity use very accurately, why does the share of renewable energy used still fall in summer?
That is a very good question.
We cross-checked this year’s actual electricity bills for months 1–8, Taipower’s wheeling data, electricity use in the same period last year and official weather data again.
The answer we ultimately saw was actually quite straightforward.
AI forecasting addresses “How much electricity will I need?”
The RE achievement rate addresses “Will enough green power be available to supply it at that time?”
These are two entirely different things.
When electricity use for summer business operations increases and air-conditioning loads rise, demand grows.
At the same time, if solar PV in the south faces prolonged rainfall or reduced sunshine, the green power actually available for wheeling also decreases.
As demand rises and supply falls, RE naturally declines.
And this time we also saw something very important:
We cannot blame everything on the weather.
Even if sunshine recovers, RE may not recover if supply scale, the generation-site portfolio and the timing allocations for individual stores are not adjusted accordingly.
So when I look at corporate green power now, I do not just ask:
“How many kWh do we buy a year?”
I care more about:
When is it needed? When is supply available? Where do the shortfalls occur? Which can be complemented by different generation sites? Which require storage for time shifting?
This is also where 15-minute data becomes truly valuable.
Not in making a dashboard look more detailed.
But in turning “forecasting, supply, matching and settlement” into one operational problem.
For KiWi, the next step for AI-EDC is not to make the forecast-accuracy figures look even better, either.
It is to put PPA, different renewable energy sources, storage and 15-minute loads into the same Energy OS.
AI looks at demand.
PPA addresses supply.
15-minute matching addresses timing.
Storage addresses mismatches.
Finally, we return to actual bills and wheeling results for verification.
This week’s biggest reminder is simple:
An accurate forecast only means we can see clearly. Real energy management also means supplying enough and matching it correctly.
Making every kWh smarter.
Steve
#KiWiNewEnergy #AIEDC #GreenPower #PPA #EnergyStorage #15MinuteMatching #SmartEnergy #MakingEverykWhSmarter
Three lines of inquiry
- Forecasting answers the question of demand; supply answers the question of availability.
- Annual procurement and matching within the same time interval must not be conflated.
- Bring supply-demand shortfalls back to bills, wheeling and attribute records for verification.
TECHNICAL EDITORIAL · AI-ASSISTED
Technical Deep Dive · AI-assisted editorial supplement
The following AI-assisted editorial supplement draws on the original text and public references. It is not Steve’s original text or new views attributed to him.
Accurate forecasting is not sufficient supply
Forecast error compares forecast and actual demand; it does not measure the amount or availability of green power that can be supplied. Even with accurate demand forecasts, insufficient supply or mismatched timing leaves shortfalls. The multi-site example, data cross-checks and observations in the original text are all the author’s account; this article has not independently verified them and adds no new measured results.
Define same-interval coverage clearly
This simplified operational metric is a load-weighted matching share, not the proportion of fully covered intervals, nor an official EnergyTag formula. First establish geographic and transmission eligibility, contractual allocation eligibility and valid attribute ownership for each site and interval; then take the lower of that site’s actual load and eligible allocated renewable energy, sum these values and divide by total load. If sharing across sites is not permitted, do not first combine supply and demand across sites before taking the lower value. Inputs must be non-negative and use consistent meter boundaries, periods, kWh and time zones; missing meter data must remain visible and coverage must be disclosed, not silently removed; the metric is undefined when the denominator is zero. Do not double count or count energy already assigned to charging toward direct load at the same time. This is not a Scope 2 compliance calculation; distinguish the annual eligible-renewable-energy numerator from the electricity-consumption denominator, and annual procurement from interval-by-interval matching. Renewable energy is also not equivalent to the broader category of carbon-free energy.
Generation portfolios require real-world conditions
Portfolio decisions should examine actual seasonal and time-interval profiles, geographic location, transmission availability and contractual conditions, rather than compare annual totals alone. Correlation between rainfall and supply changes is not sufficient on its own to establish quantified causality; potential factors such as weather, capacity, dispatch and load allocation must be distinguished. The direction outlined in the original text can serve as a starting point for examination, not as a conclusion verified by this draft.
Storage shifts time, but retains constraints
Storage can only shift energy that is already available, and entails losses and limits on charging/discharging power, capacity and state of charge (SOC). Grid charging does not automatically become renewable energy; only eligible discharge quantities whose charging-source attributes are traceable, after deducting losses, may count as renewable supply. Charging energy must not simultaneously count toward end-use electricity matching and subsequent discharge matching, and energy or attributes must not be allocated twice. Assessment must use deliverable quantities, not idealized lossless shifted quantities.
Time granularity is not a compliance guarantee
15 minutes can be the time resolution for operational analysis, but does not itself prove compliance or RE100 conformity, or guarantee cost savings or green power outcomes. EnergyTag’s hourly schemes and annual accounting serve different purposes; citing technical standards is not a statement that KiWi has obtained certification. PPA supply obligations, wheeling and settlement remain subject to applicable contracts and rules; Scope 2 must be assessed separately under applicable guidance, and analytical granularity cannot substitute for accounting requirements.
A verification loop makes shortfalls visible
Using consistent meters, periods, units and time zones, progressively compare forecasts, actual demand, measurements, wheeling, bills and attribute records. Clearly identify missing data, energy that cannot be allocated and unmatched load, then feed this back into generation-portfolio and operational planning. Without these original records, specific coverage or cost savings cannot be claimed; this diagram shows only the relationships to examine, not a deployed process or actual results.
References · Read 2026-10-07
- EnergyTag FAQ
Explains differences between hourly and annual accounting, as well as considerations for geography, traceability and avoiding double counting; storage must reflect losses. It provides background on a voluntary standard, not a basis for mandatory 15-minute compliance.
- EnergyTag Granular Certificate Scheme Standard V2
The Scheme Standard for granular certificates, not the separately developed Matching Standard. It serves only as background on the boundaries of time and energy-attribute records, not as the official source of this article’s operational formula; citation does not mean that KiWi, this process or any site has been certified.
- GHG Protocol · Scope 2 Guidance
Official accounting guidance and updates for purchased-electricity emissions. Operational matching metrics are not equivalent to Scope 2 accounting; this draft does not describe 15-minute matching as a universal mandatory requirement.